Doing Better at Doing Good: Lessons from a Bayesian Meta-Analysis After 20 Years of Poverty Graduation Programs

Multifaceted poverty Graduation programs that aim to increase income generating activities for low-income households are being rapidly adopted and scaled by nonprofit organizations and governments. This paper conducts a Bayesian hierarchical meta-analysis of 49 studies (encompassing 81 treatment variations) on poverty Graduation programs, defined as containing at least three of five components: consumption support, savings facilitation, skills training, coaching, and a lump-sum transfer. The analysis finds large gains across economic and psychosocial outcomes: monthly consumption rises by 38 USD PPP (11 percent), income by 28 USD PPP (15 percent), alongside improvements in savings, assets, food security, mental health, and women's empowerment. Beyond these average treatment effects, three results stand out. First, programs with large transfers (at least 400 USD PPP) have larger treatment effects but are not more cost-effective. Second, more expensive programs generate larger treatment effects, but the improvements are not commensurate with the cost increases; thus, lower cost programs tend to be more cost-effective. Third, programs implemented from 2016 onward perform considerably better than those implemented prior to 2016: benefit-cost ratios from 2016 onward are 5.39, compared to 2.89 through 2015. Neither changes in cost nor in high-level design (components) explain this upward trend. Considering other heavily studied interventions, the results show a similar upward trend for treatment effects on behavior change from financial education but not for unconditional cash transfers. The paper hypothesizes a mechanism of continuous sector-community learning and improved implementation quality.